Feasibility of Pretreatment FDG PET Radiomics in Predicting Complete Pathological Response after Neoadjuvant Concurrent Chemoradiotherapy for Esophageal Cancer

Abstract

<h3>Purpose/Objective(s)</h3> To investigate the feasibility in predicting the pathological complete pathological response (pCR) of esophageal tumors after neoadjuvant concurrent chemoradiotherapy (CCRT) using pretreatment 18-fluorodeoxyglucose positron emission tomography (FDG PET) images. <h3>Materials/Methods</h3> This study retrospectively included 60 esophageal tumors of 59 patients from November 2012 to August 2021. All patients received neoadjuvant CCRT followed by radical surgery. The FDG PET examination before CCRT was used for evaluation. First, the maximum standardized uptake value (SUVmax) within a tumor was identified and then utilized to form a metabolic tumor volume (MTV) by grouping a set of voxels connecting to the SUVmax and with SUV not less than SUVmax * 0.5. Next, the texture indices were further calculated to describe the heterogeneity of the FDG uptakes discretized as gray level (GL) within MTV. Finally, receiver operating characteristic (ROC) curves analysis assessed the predictive performance of each index. In the analysis, pCR was defined as a positive event. <h3>Results</h3> Fifty-seven out of 59 patients were men (57/59, 96.6%). The number of tumors located at the upper, middle, and lower third of the esophagus were 9, 12, and 39, respectively. Ninety-five percent of the tumors were squamous cell carcinomas (57/60, 95%). Seventeen tumors of 17 patients were proved as pCR (17/60, 28.3%). The Chi-square test demonstrated that the associations between treatment response (pCR and non-pCR) and tumor location, clinical T, and N stage were insignificant (p=.16, .193, and .254). The SUVmax demonstrated a significant ability in identifying pCR tumors (AUC=.694, p=.013), whereas MTV was nonsignificant. The tumors with high metabolic uptake corresponded to a high probability of pCR. The texture indices, including GL nonuniformity normalized, low GL zone emphasis, and small area low GL emphasis of zones, GL nonuniformity normalized, long run low GL emphasis, low GL run emphasis, and short run low GL emphasis of runs, and low GL emphasis and small dependence low GL emphasis of GL dependencies, exhibited a significant ability to differentiate pCR from non-pCR tumors (AUC=.331, .289, .298, .339, .311, .311, .316, .312, and .302). These texture indices reflected that a low evaluation corresponded to a high likelihood of pCR. The tumors with a low or low uptake associated with small areas demonstrated a high probability of non-pCR. <h3>Conclusion</h3> This study revealed that pCR for esophageal cancer tumors could be predicted using metabolic and texture indices. Future studies are warranted for clinical application.